> ## Documentation Index
> Fetch the complete documentation index at: https://docs.protectorplus.cloudsine.tech/llms.txt
> Use this file to discover all available pages before exploring further.

# Detection Layers

> Standard Guardrails and ShieldPrompt — two complementary layers in defence-in-depth.

Protector Plus combines two complementary layers in defence-in-depth: deterministic Standard Guardrails for known patterns, and ShieldPrompt for model-driven detection of evasive and emerging threats.

## Standard Guardrails — rule-based, CPU-powered

Deterministic detection for well-known risk vectors. CPU-only, suitable for high throughput.

| Guardrail                  | Phase          | Description                                                                                         |
| -------------------------- | -------------- | --------------------------------------------------------------------------------------------------- |
| **Keyword Blocklist**      | Input + Output | Exact keyword/phrase blocklist on prompts and outputs.                                              |
| **Regex Pattern Matching** | Input + Output | Regular-expression filtering against operator-defined patterns.                                     |
| **PII Detection**          | Input + Output | NER-based detection: names, emails, phone numbers, Singapore NRIC/FIN, passport, credit cards, etc. |

## ShieldPrompt — defence-in-depth orchestration, GPU-powered

The orchestration pipeline that combines model-based detection layers.

| Layer                        | Phase          | Role                                                                                                           |
| ---------------------------- | -------------- | -------------------------------------------------------------------------------------------------------------- |
| **LLM Classifier**           | Input + Output | LLM-as-judge that scores prompt-injection likelihood and threat intent.                                        |
| **Vector Filter / TVDB**     | Input          | Semantic similarity against the proprietary Threat Vector Database. Catches paraphrased and rephrased attacks. |
| **Content Moderation**       | Output         | LLM-based classifier for hazardous content (violence, hate, unethical, etc.) in LLM responses.                 |
| **System Prompt Protection** | Output         | LLM-based detection of system-prompt leakage in LLM responses.                                                 |

See [ShieldPrompt](/architecture/shieldprompt) for the deep dive.

## Block vs Monitor

Each guardrail (in inline deployments) can operate in **Block** or **Monitor** mode. Block drops the connection on violation; Monitor passes the prompt through but records the detection on the Alerts page.